[PATCH 0/2] use blk_rq_nr_phys_segments() instead of iterating bvecs

[PATCH 0/2] use blk_rq_nr_phys_segments() instead of iterating bvecs

[Caleb Sander Mateos]

Minor simplification to loop and zloop to get the number of segments
from the request directly instead of iterating over all its bvecs.

Caleb Sander Mateos (2):
  loop: use blk_rq_nr_phys_segments() instead of iterating bvecs
  zloop: use blk_rq_nr_phys_segments() instead of iterating bvecs

 drivers/block/loop.c  | 5 +----
 drivers/block/zloop.c | 5 +----
 2 files changed, 2 insertions(+), 8 deletions(-)

-- 
2.45.2

[PATCH 1/2] loop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Caleb Sander Mateos]

The number of bvecs can be obtained directly from struct request's
nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
instead of iterating over the bvecs an extra time.

Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
---
 drivers/block/loop.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 13ce229d450c..8096478fad45 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c
@@ -346,16 +346,13 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
 	struct request *rq = blk_mq_rq_from_pdu(cmd);
 	struct bio *bio = rq->bio;
 	struct file *file = lo->lo_backing_file;
 	struct bio_vec tmp;
 	unsigned int offset;
-	int nr_bvec = 0;
+	unsigned short nr_bvec = blk_rq_nr_phys_segments(rq);
 	int ret;
 
-	rq_for_each_bvec(tmp, rq, rq_iter)
-		nr_bvec++;
-
 	if (rq->bio != rq->biotail) {
 
 		bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
 				     GFP_NOIO);
 		if (!bvec)
-- 
2.45.2

[PATCH 2/2] zloop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Caleb Sander Mateos]

The number of bvecs can be obtained directly from struct request's
nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
instead of iterating over the bvecs an extra time.

Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
---
 drivers/block/zloop.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/drivers/block/zloop.c b/drivers/block/zloop.c
index 92be9f0af00a..7883e56ed12a 100644
--- a/drivers/block/zloop.c
+++ b/drivers/block/zloop.c
@@ -368,11 +368,11 @@ static void zloop_rw(struct zloop_cmd *cmd)
 	struct req_iterator rq_iter;
 	struct zloop_zone *zone;
 	struct iov_iter iter;
 	struct bio_vec tmp;
 	sector_t zone_end;
-	int nr_bvec = 0;
+	unsigned short nr_bvec = blk_rq_nr_phys_segments(rq);
 	int ret;
 
 	atomic_set(&cmd->ref, 2);
 	cmd->sector = sector;
 	cmd->nr_sectors = nr_sectors;
@@ -435,13 +435,10 @@ static void zloop_rw(struct zloop_cmd *cmd)
 		zone->wp += nr_sectors;
 		if (zone->wp == zone_end)
 			zone->cond = BLK_ZONE_COND_FULL;
 	}
 
-	rq_for_each_bvec(tmp, rq, rq_iter)
-		nr_bvec++;
-
 	if (rq->bio != rq->biotail) {
 		struct bio_vec *bvec;
 
 		cmd->bvec = kmalloc_array(nr_bvec, sizeof(*cmd->bvec), GFP_NOIO);
 		if (!cmd->bvec) {
-- 
2.45.2

Re: [PATCH 1/2] loop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Ming Lei]

On Sat, Nov 08, 2025 at 04:01:00PM -0700, Caleb Sander Mateos wrote:
> The number of bvecs can be obtained directly from struct request's
> nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
> instead of iterating over the bvecs an extra time.
> 
> Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
> ---
>  drivers/block/loop.c | 5 +----
>  1 file changed, 1 insertion(+), 4 deletions(-)
> 
> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> index 13ce229d450c..8096478fad45 100644
> --- a/drivers/block/loop.c
> +++ b/drivers/block/loop.c
> @@ -346,16 +346,13 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
>  	struct request *rq = blk_mq_rq_from_pdu(cmd);
>  	struct bio *bio = rq->bio;
>  	struct file *file = lo->lo_backing_file;
>  	struct bio_vec tmp;
>  	unsigned int offset;
> -	int nr_bvec = 0;
> +	unsigned short nr_bvec = blk_rq_nr_phys_segments(rq);
>  	int ret;
>  
> -	rq_for_each_bvec(tmp, rq, rq_iter)
> -		nr_bvec++;
> -

The two may not be same, since one bvec can be splitted into multiple segments.

Thanks,
Ming

Re: [PATCH 1/2] loop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Chaitanya Kulkarni]

On 11/9/25 04:19, Ming Lei wrote:
> On Sat, Nov 08, 2025 at 04:01:00PM -0700, Caleb Sander Mateos wrote:
>> The number of bvecs can be obtained directly from struct request's
>> nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
>> instead of iterating over the bvecs an extra time.
>>
>> Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
>> ---
>>   drivers/block/loop.c | 5 +----
>>   1 file changed, 1 insertion(+), 4 deletions(-)
>>
>> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
>> index 13ce229d450c..8096478fad45 100644
>> --- a/drivers/block/loop.c
>> +++ b/drivers/block/loop.c
>> @@ -346,16 +346,13 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
>>   	struct request *rq = blk_mq_rq_from_pdu(cmd);
>>   	struct bio *bio = rq->bio;
>>   	struct file *file = lo->lo_backing_file;
>>   	struct bio_vec tmp;
>>   	unsigned int offset;
>> -	int nr_bvec = 0;
>> +	unsigned short nr_bvec = blk_rq_nr_phys_segments(rq);
>>   	int ret;
>>   
>> -	rq_for_each_bvec(tmp, rq, rq_iter)
>> -		nr_bvec++;
>> -
> The two may not be same, since one bvec can be splitted into multiple segments.
>
> Thanks,
> Ming
>
>

I had this patch only to realize this :-

bvec_split_segs():-
* When splitting a bio, it can happen that a bvec is encountered that is too
* big to fit in a single segment and hence that it has to be split in the
* middle. This function verifies whether or not that should happen. The 
value
* %true is returned if and only if appending the entire @bv to a bio with
* *@nsegs segments and *@sectors sectors would make that bio 
unacceptable for
* the block driver.

if following messup the indentation after I send plz ignore :-

submit_bio()
   -> submit_bio_noacct()
     -> __submit_bio_noacct()
       -> __submit_bio()
         -> blk_mq_submit_bio()
           -> __bio_split_to_limits()
             -> bio_split_rw()
               -> bio_split_rw_at()
                 -> bio_split_io_at()

                   -- bio_for_each_bvec()   (iterate all bvecs)
                    -- [Fast path]
                         nsegs++             (1 bvec = 1 segment)
                         bytes += bv.bv_len
                    -- [Slow path]
                       /* one bv is passed from bio bvec list  */
                       -> bvec_split_segs(lim, &bv, &nsegs, &bytes,
                                          lim->max_segments, max_bytes)

                            /* working on one bvec and associated 
segments */
                            -- while (len && *nsegs < max_segs) {

                                 ...

                                 get_max_segment_size()
                                   -> min(len,
                                          seg_boundary_mask - offset,
                                          max_segment_size)

                                 (*nsegs)++
                                 total_len += seg_size
                                 len -= seg_size

                                 ...

                               }

                            -- *bytes += total_len
                            -- return (len > 0)

           -> blk_mq_bio_to_request()
                rq->nr_phys_segments = nr_segs

-ck

Re: [PATCH 1/2] loop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Caleb Sander Mateos]

On Sun, Nov 9, 2025 at 4:20 AM Ming Lei <ming.lei@redhat.com> wrote:
>
> On Sat, Nov 08, 2025 at 04:01:00PM -0700, Caleb Sander Mateos wrote:
> > The number of bvecs can be obtained directly from struct request's
> > nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
> > instead of iterating over the bvecs an extra time.
> >
> > Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
> > ---
> >  drivers/block/loop.c | 5 +----
> >  1 file changed, 1 insertion(+), 4 deletions(-)
> >
> > diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> > index 13ce229d450c..8096478fad45 100644
> > --- a/drivers/block/loop.c
> > +++ b/drivers/block/loop.c
> > @@ -346,16 +346,13 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
> >       struct request *rq = blk_mq_rq_from_pdu(cmd);
> >       struct bio *bio = rq->bio;
> >       struct file *file = lo->lo_backing_file;
> >       struct bio_vec tmp;
> >       unsigned int offset;
> > -     int nr_bvec = 0;
> > +     unsigned short nr_bvec = blk_rq_nr_phys_segments(rq);
> >       int ret;
> >
> > -     rq_for_each_bvec(tmp, rq, rq_iter)
> > -             nr_bvec++;
> > -
>
> The two may not be same, since one bvec can be splitted into multiple segments.

Hmm, io_buffer_register_bvec() already assumes
blk_rq_nr_phys_segments() returns the number of bvecs iterated by
rq_for_each_bvec(). I asked about this on the patch adding it, but
Keith assures me they match:
https://lore.kernel.org/io-uring/Z7TmrB4_aBnZdFbo@kbusch-mbp/.

Best,
Caleb

Re: [PATCH 1/2] loop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Chaitanya Kulkarni]

On 11/10/25 08:47, Caleb Sander Mateos wrote:
> On Sun, Nov 9, 2025 at 4:20 AM Ming Lei <ming.lei@redhat.com> wrote:
>> On Sat, Nov 08, 2025 at 04:01:00PM -0700, Caleb Sander Mateos wrote:
>>> The number of bvecs can be obtained directly from struct request's
>>> nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
>>> instead of iterating over the bvecs an extra time.
>>>
>>> Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
>>> ---
>>>   drivers/block/loop.c | 5 +----
>>>   1 file changed, 1 insertion(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/block/loop.c b/drivers/block/loop.c
>>> index 13ce229d450c..8096478fad45 100644
>>> --- a/drivers/block/loop.c
>>> +++ b/drivers/block/loop.c
>>> @@ -346,16 +346,13 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
>>>        struct request *rq = blk_mq_rq_from_pdu(cmd);
>>>        struct bio *bio = rq->bio;
>>>        struct file *file = lo->lo_backing_file;
>>>        struct bio_vec tmp;
>>>        unsigned int offset;
>>> -     int nr_bvec = 0;
>>> +     unsigned short nr_bvec = blk_rq_nr_phys_segments(rq);
>>>        int ret;
>>>
>>> -     rq_for_each_bvec(tmp, rq, rq_iter)
>>> -             nr_bvec++;
>>> -
>> The two may not be same, since one bvec can be splitted into multiple segments.
> Hmm, io_buffer_register_bvec() already assumes
> blk_rq_nr_phys_segments() returns the number of bvecs iterated by
> rq_for_each_bvec(). I asked about this on the patch adding it, but
> Keith assures me they match:
> https://lore.kernel.org/io-uring/Z7TmrB4_aBnZdFbo@kbusch-mbp/.
>
> Best,
> Caleb
>
Perhaps I don't understand how they will be same ? can share more details.
Segment Splitting :-

nr_bvec=1, blk_rq_nr_phys_segments=2 (see below)
- ONE large bvec split into MULTIPLE physical segments
- Patch above allocate array[2], but iterate and fill only array[0] ?

*[ 6155.673749] nullb_bio: 128K bio as ONE bvec: sector=0, size=131072, op=WRITE*
*[ 6155.673846] null_blk: #### null_handle_data_transfer:1375*
*[ 6155.673850] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=2*
*[ 6155.674263] null_blk: #### null_handle_data_transfer:1375*
*[ 6155.674267] null_blk: nr_bvec=1 blk_rq_nr_phys_segments=1*

diff --git a/drivers/block/null_blk/main.c b/drivers/block/null_blk/main.c
index 1fe3373431ca..74ab0ba53f62 100644
--- a/drivers/block/null_blk/main.c
+++ b/drivers/block/null_blk/main.c
@@ -1364,7 +1364,18 @@ static blk_status_t null_handle_data_transfer(struct nullb_cmd *cmd,
         unsigned int max_bytes = nr_sectors << SECTOR_SHIFT;
         unsigned int transferred_bytes = 0;
         struct req_iterator iter;
+       struct req_iterator rq_iter;
         struct bio_vec bvec;
+       int nr_bvec = 0;
+
+       rq_for_each_bvec(bvec, rq, rq_iter)
+               nr_bvec++;
+
+       if (req_op(rq) == REQ_OP_WRITE) {
+               pr_info("#### %s:%d\n", __func__, __LINE__);
+               pr_info("nr_bvec=%d blk_rq_nr_phys_segments=%u\n",
+                       nr_bvec, blk_rq_nr_phys_segments(rq));
+       }
  
         spin_lock_irq(&nullb->lock);
         rq_for_each_segment(bvec, rq, iter) {

-ck

Re: [PATCH 1/2] loop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Ming Lei]

On Mon, Nov 10, 2025 at 08:47:46AM -0800, Caleb Sander Mateos wrote:
> On Sun, Nov 9, 2025 at 4:20 AM Ming Lei <ming.lei@redhat.com> wrote:
> >
> > On Sat, Nov 08, 2025 at 04:01:00PM -0700, Caleb Sander Mateos wrote:
> > > The number of bvecs can be obtained directly from struct request's
> > > nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
> > > instead of iterating over the bvecs an extra time.
> > >
> > > Signed-off-by: Caleb Sander Mateos <csander@purestorage.com>
> > > ---
> > >  drivers/block/loop.c | 5 +----
> > >  1 file changed, 1 insertion(+), 4 deletions(-)
> > >
> > > diff --git a/drivers/block/loop.c b/drivers/block/loop.c
> > > index 13ce229d450c..8096478fad45 100644
> > > --- a/drivers/block/loop.c
> > > +++ b/drivers/block/loop.c
> > > @@ -346,16 +346,13 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
> > >       struct request *rq = blk_mq_rq_from_pdu(cmd);
> > >       struct bio *bio = rq->bio;
> > >       struct file *file = lo->lo_backing_file;
> > >       struct bio_vec tmp;
> > >       unsigned int offset;
> > > -     int nr_bvec = 0;
> > > +     unsigned short nr_bvec = blk_rq_nr_phys_segments(rq);
> > >       int ret;
> > >
> > > -     rq_for_each_bvec(tmp, rq, rq_iter)
> > > -             nr_bvec++;
> > > -
> >
> > The two may not be same, since one bvec can be splitted into multiple segments.
> 
> Hmm, io_buffer_register_bvec() already assumes
> blk_rq_nr_phys_segments() returns the number of bvecs iterated by
> rq_for_each_bvec(). I asked about this on the patch adding it, but
> Keith assures me they match:
> https://lore.kernel.org/io-uring/Z7TmrB4_aBnZdFbo@kbusch-mbp/.

blk_rq_nr_phys_segments() >= nr_bvec, would you like to cook a fix?

Thanks, 
Ming

Re: [PATCH 2/2] zloop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Christoph Hellwig]

On Sat, Nov 08, 2025 at 04:01:01PM -0700, Caleb Sander Mateos wrote:
> The number of bvecs can be obtained directly from struct request's
> nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
> instead of iterating over the bvecs an extra time.

Same reason this doesn't work as Ming explained for ublk.

Maybe we should lift this code from loop/zloop into a well documented
common helper to make it more obvious?

Re: [PATCH 2/2] zloop: use blk_rq_nr_phys_segments() instead of iterating bvecs

[Chaitanya Kulkarni]

On 11/11/25 01:07, Christoph Hellwig wrote:
> On Sat, Nov 08, 2025 at 04:01:01PM -0700, Caleb Sander Mateos wrote:
>> The number of bvecs can be obtained directly from struct request's
>> nr_phys_segments field via blk_rq_nr_phys_segments(), so use that
>> instead of iterating over the bvecs an extra time.
> Same reason this doesn't work as Ming explained for ublk.
>
> Maybe we should lift this code from loop/zloop into a well documented
> common helper to make it more obvious?
>
>
Absolutely, patch sent with added quantitative data from my experiments,
to prove why above is wrong, please have a look.

-ck